An experiment on corruption and gender

M. Fernanda Rivas�y

December 2008

Abstract

There exists evidence in the social science literature that women may be more relationship-

oriented, may have higher standards of ethical behavior and may be more concerned with

the common good than men are. This would imply that women are more willing to sacri�ce

private pro�t for the public good, which would be especially important for political life. A

number of papers with �eld data have found di¤erences in the corrupt activities of males

and females, nonetheless they have drawbacks that may be overcome in a lab experiment.

The aim of this paper is to see experimentally if women and men, facing the same situation

behave in a di¤erent way, as suggested in the �eld-data studies or, on the contrary, they

behave in the same way. The results found in the experiment show that women are indeed

less corrupt than men.

Keywords: corruption, gender, experiment

JEL classi�cation: C91, D73, J16

�University of Granada. Email: mfrivas@ugr.esyI thank Jordi Brandts for his supervision, and Selim Ergun for his help in running the experiments and for

his helpful comments.

1

1 Introduction

A large stream of research papers has documented systematic gender di¤erences in behavior.

This evidence suggests that women may be more relationship-oriented, may have higher stan-

dards of ethical behavior and may be more concerned with the common good than men are.

This would imply that women are more willing to sacri�ce private pro�t for the public good,

and this would be especially important for political life.

Criminology literature shows that men are more likely to commit o¤enses than females

(Gottfredson and Hirschi, 1990). This gender di¤erence does not vary over time and countries;

it starts at low ages and is maintained through the years. Additionally, they show evidence that

crimes committed by men are more serious than those committed by women.

In the economics literature, a number of �eld data papers have found that women are less

tolerant toward dishonest behavior, and that there exists a negative relation between women�s

participation rate in politics and corruption level. However, it can be argued that the observed

gender di¤erence in behavior in the �eld may be due to other reasons than real gender di¤erences

in corrupt behavior; possible interferences that can be controlled in a laboratory experiment. One

possible reason is the di¤erent degree of risk aversion of males and females as observed in many

experiments. This could be an important determinant of gender di¤erence given that corrupt

behavior (almost) always implies a probability of being discovered and thus being punished.

Therefore, a gender di¤erence in behavior could be due to a di¤erence in risk aversion � women

who are more risk averse would avoid risky activities such as getting involved in corrupt activities

more frequently than men. In this experiment I control for risk aversion, trying to isolate the

results from the in�uence of this variable.

Another possible explanation for the observed gender di¤erence is that women have entered

the labor market and politics much more recently than men. The di¤erence may therefore be

due to di¤erences in terms of accessing networks of corruption, or in terms of knowledge about

how to get involved in corrupt activities. Thus, it may be just a matter of time until women

get involved in corrupt activities. In my experiment, subjects do not have a previous history,

they do not have experience in this kind of game. Therefore, this e¤ect � the di¤erences in

experience � can be ruled out as an explanation of any di¤erence observed in the lab.

A third feasible explanation is that low levels of corruption and high female political partic-

ipation are both the result of a liberal democracy that simultaneously promotes gender equality

2

and good governance. This limitation of the �eld studies can be overcome in a laboratory

experiment where we observe male and female decisions taken in the same environment.

This paper addresses the question of whether women and men facing the same situation

behave in a di¤erent way (as suggested in the �eld data papers), or on the contrary, whether

they behave similarly when women are in the same position as men. As explained in the

Handbook on Fighting Corruption (Phyllis and Kpundeh, 1999), "In broad terms, corruption

is the abuse of public o¢ ce for private gain". In this paper, corrupt behavior refers to bribery.

Bribery is a form of pecuniary corruption; it is an act usually involving the giving of money or

gifts to in�uence the behavior of the recipient in ways that are inconsistent with that person�s

duties or in violation of the law. In the experiment, I observe bribery through the manipulation

of public o¢ cials�decisions in exchange for bribes.1 The experimental design tries to capture

the characteristics of corrupt behavior: the reciprocity between the briber and the o¢ cial, the

negative externality over the public, and the probability of being discovered.

In the experiment there are two types of players: �rms and public o¢ cials. The possibility

of corruption is introduced by allowing the �rm to send some amount of money as a bribe to the

public o¢ cial, who has to decide between two alternatives. One of the alternatives is slightly

better for the o¢ cial, while the other is much more favorable to the �rm, but has negative

externalities over all the other players. I conduct four sessions: two sessions where subjects of

both genders participate (one gender in one role and the other gender in the other role) and two

sessions with only one gender participating. The objective is to check whether or not men and

women behave di¤erently according to their partner�s gender. If women are stereotyped as less

corrupt, I should observe that �rms o¤er a bribe to female public o¢ cials less frequently than to

male o¢ cials. Therefore, I also ask �rms about their expectations regarding the corruptibility

of public o¢ cials.

The results show that if the �rm is a woman, even when controlling for relevant variables,

a lower bribe is o¤ered. When asked about their beliefs regarding the probability that a public

o¢ cial will accept the bribe and choose the most favorable alternative for the �rm � but with

negative consequences for the other participants� both male and female �rms expect female

public o¢ cials to accept the bribe and choose the corrupt alternative less frequently than male

o¢ cials. Nonetheless, the econometric analysis shows that the partner�s gender is not statistically

1 In what follows I call the manipulation of public o¢ cials�decisions in exchange for bribes either bribery or

corruption.

3

signi�cant. Concerning the behavior of public o¢ cials, I �nd that if the public o¢ cial is a woman

and the �rm is also a woman, the probability of accepting the bribe is lower and the probability

of choosing the corrupt alternative is also lower.

The di¤erence in the assigned probabilities would suggest some stereotyping behavior in that

both genders expect women to be less corrupt than men. On the other hand, when analyzing

their actual behavior, they do not show overall signi�cant di¤erences according to the gender of

their partners.

The main result of the experiment is that women are less corrupt than men. Therefore, the

conclusion is in line with the �eld data papers. It can thus be expected that increasing female

participation in the labor force and politics would help to �ght corruption.

The paper is organized as follows: in the next subsection �eld data papers dealing with

gender and corruption are reviewed. An overview of some papers on experiments on gender

di¤erences and corruption are given in Subsections 1.2 and 1.3, respectively. The experimental

design is explained in Section 2, the results are presented in Section 3, and �nally conclusions

are drawn in Section 4.

1.1 Literature on corruption and gender with �eld data

Dollar et al. (2001), Swamy et al. (2001), and Torgler and Valev (2006) show empirically that

men are more frequently involved in corrupt activities than women. These papers study the

acceptability of corrupt behavior by men and women and the relationship between women�s

participation rate in politics and a corruption index.

Analyzing data for more than 100 countries, Dollar et al. (2001) �nd a strong negative

and statistically signi�cant relationship between the level of female participation in politics �

measured by the percentage of seats occupied by women in the lower and upper chambers �

and a corruption index � the International Country Risk Guide index. Their conclusion is

that encouraging women to have a higher political participation may be bene�cial to the whole

society.

Using di¤erent data sources, Swamy et al. (2001) study the hypothesis that female partici-

pation in the government would reduce corruption. Using the World Value Surveys (WVS) they

�nd that women are less tolerant towards dishonest or illegal activities than men. The second

data set they use is an enterprise survey conducted in Georgia, where �rm owners and managers

4

were asked the following question: "How frequently do the o¢ cials providing the service require

uno¢ cial payments?". The evidence they �nd suggests that men are more frequently involved

in corruption than women. With the third data set using cross-country data 2, they �nd that

corruption is less prevalent where women carry more weight in politics and in the labor force,

in line with Dollar et al. (2001).

In a more recent paper, Torgler and Valev (2006) investigate empirically if women are more

willing to be compliant than men. Analyzing the WVS and the European Values Survey, they

�nd a strong gender e¤ect: women are more willing to comply than men, and are less likely to

agree that corruption and cheating on taxes can be justi�ed. Moreover, they do not observe

a decline over time in the gender di¤erence, thus it contradicts the role theory which suggests

that greater equality in status between men and women would decrease gender di¤erences over

time.3

Although these papers control for variables other than gender that may explain the corrup-

tion index or the acceptability of corruption (e.g. age, education, religion, earnings, etc.), they

still have some drawbacks. The main criticism of these papers is that gender di¤erence may

be driven by the existence of a liberal democracy that simultaneously promotes a high level of

gender equality � and therefore a high level of female participation in the labor market and

politics� and low levels of corruption. If this were the case, then a relationship would exist

between low corruption indexes and high levels of female participation in politics, but a higher

female participation in politics would not cause lower levels of corruption. Instead, both things

would be caused by the presence of a liberal democracy. This potential problem is ruled out

in the experiment. The subjects in the experiment � both men and women � are exposed to

the same situation where they can freely choose what to do. If we �nd a gender di¤erence after

controlling for the appropriate variable, it should be due to a real di¤erence in behavior between

men and women.

2They use the Transparency International Corruption Perception Index, which is based on di¤erent information

sources such as investor surveys and assessments of country experts.3Criminology literature also suggests that the role theory cannot explain di¤erences in crime rates between

men and women (Gottfredson and Hirschi, 1990).

5

1.2 Experiments on gender di¤erences

Many papers study gender di¤erences in behavior, although they yield contradictory results.

Some papers �nd di¤erences and others do not. In what follows, I review some papers that

analyze issues related to corruption such as risk aversion, giving, reciprocity, and deception.

As regards risk aversion, Croson and Gneezy (2008) review evidence that shows that men

are more risk taking than women. The observed di¤erences in risk perception (men are more

overcon�dent than women) may explain, in part, the di¤erences in risk taking. A number of

papers reviewed in Croson and Gneezy (2008) support the evolutionary argument rather than

the socialization argument to explain the gender di¤erence.4 Nevertheless, some papers �nd

di¤erent results. Schubert et al. (1999) �nd that in a strongly ambiguous context women are

more risk averse when lottery choices are framed as gains, while men are more risk averse when

lotteries are framed as losses. They �nd no gender di¤erences when the context is abstract

or weakly ambiguous. Johnson and Powell (1994) �nd that gender di¤erences depend on the

subject pool. In the subpopulation of managers, for example, they do not �nd di¤erences in risk

taking. The absence of a di¤erence may be due to self selection, or adaptation to the demands of

the job. Eckel and Grossman (2002) analyze whether there exist gender di¤erences in �nancial

risk and also measure gender di¤erences in expectations about other�s behavior. Both genders

estimate correctly that women have a higher degree of risk aversion than men. However, they

overestimate the degree of risk aversion � especially of women. Eckel and Grossman argue that

if women are stereotyped as more risk averse, this could a¤ect them negatively in many aspects

ranging from lower wages to less aggressive health treatments.

As concerns o¤ers and acceptance/rejection, Eckel and Grossman (2001) �nd no gender dif-

ference in an ultimatum game in the proposer behavior, but in the responder rate of rejections:

women are more likely to accept lower o¤ers. In a solidarity game � a variation of the dictator

game � Selten and Ockenfels (1998) �nd that women are more generous (or show more soli-

darity) than men. In terms of reciprocity, a number of papers �nd no gender di¤erences (Eckel

and Wilson, 2004a; Eckel and Wilson, 2004b), while others �nd that women are more reciprocal

4Evolutionary theories hold that sex di¤erences are dependent on reproduction: men are more risk taking in

the period when they are trying to attract mates, while women are more risk averse in the child-bearing period

(Wood and Eagly, 2002). The socialization theory is not wholly focused on biology, but on cultural and social

practices. Because men and women tend to have di¤erent social roles, they become psychologically di¤erent to

adjust to their social roles (Eagly and Wood, 1999).

6

than men (Chaudhuri and Gangadharan, 2003; Snijdes and Keren, 2004).

Croson and Gneezy (2008) explain all these di¤erences by di¤erent reactions to the context.

Most of the di¤erences mentioned above are due to changes in female behavior according to the

context and how the experiment is framed, rather than changes in male behavior.

Another paper in this line is that of Dreber and Johannesson (2008). They study gender

di¤erences in deception in a sender-receiver game, �nding that male senders are more prone

than female senders to lie in order to secure a higher payo¤ to the detriment of their partners�

payo¤ . Although the topics � deception and corruption � are not the same, they are related

in that both imply negative externalities to other subjects.

This paper contributes to this literature by experimentally examining whether there exist

di¤erences in the corrupt behavior of women and men.

1.3 Experiments on corruption

Given the di¢ culty of collecting reliable �eld data on corrupt activities due to the secrecy in

which they take place, in recent years the topic has been studied using laboratory experiments.

In this subsection I review three research papers on corruption using experiments, Abbink et

al. (2002), Abbink and Hennig-Schmidt (2002), and Alatas et al. (forthcoming). The �rst two

papers deal with bribery games � whose design I follow in this paper � while the last paper

studies gender di¤erences in the acceptability of corruption.

Abbink et al. (2002) present the �rst interactive experimental corruption game.5 They

model bribery as a situation with "negative" reciprocity, implementing a two-player sequential

game where the �rst player � the potential briber � is interpreted as a businessman or a �rm

and the second player as a public o¢ cial. The �rst player can send some amount of money to

the second player in the hope of persuading him/her to make a decision favorable to the �rm.

This experiment has three di¤erent treatments to separate three characteristics of corruption:

reciprocity (between the subjects involved in the activity), negative externalities over others,

and the risk of being caught.6 In the baseline � a pure reciprocity game � they �nd that the

5Several papers look for factors that in�uence people�s corruptibility in non-interactive games, e.g. Frank and

Schulze (2000), Schulze and Frank (2003).6The corrupt relation should be based on trust and reciprocity between the subjects involved given that no

binding contract is possible. As shown in many di¤erent studies, corruption implies a negative e¤ect over the

public and implies a probability � generally small� of being discovered.

7

(non-desirable) relationship can be established through trust and reciprocity. They add negative

externalities in their second treatment, �nding no evidence of any e¤ect on decision making. In

the third treatment, which includes an external risk, they �nd less reciprocal cooperation. The

results suggest that harsh, low-probability punishment of corruption may be very preventive. In

their experiment, the instructions were written in neutral terms. In a follow-up paper, Abbink

and Hennig-Schmidt (2002) analyze if there exist di¤erences in behavior when the instructions

are loaded.7 They do not �nd signi�cant di¤erences between the neutral-instruction and the

loaded-instruction treatments, concluding that the game is rather insensitive to the way it is

presented to subjects.

Alatas et al. (forthcoming) conduct an experiment to investigate whether there exist gender

di¤erences in the acceptability of corruption and whether they di¤er between countries. They

conduct the experiment in Australia, India, Indonesia, and Singapore. The experiment is a

one-shot game where subjects play in groups of three: one �rm that can o¤er a bribe, one public

o¢ cial that can accept/reject the bribe, and one citizen that can punish the other two players.

They only �nd gender di¤erences in Australia, concluding that gender di¤erences are culture

speci�c. Although Alatas et al. (forthcoming) investigate a subject similar to that examined

in this paper, there are important di¤erences. First of all, they focus on the acceptability of

corruption and not on corrupt behavior in itself, given that the public o¢ cial does not have

more discretionary power other than accepting or rejecting the bribe. Secondly, they conduct

a one-shot experiment, whereas I conduct a 20-round experiment. I am interested in studying

corrupt behavior in a long-run relationship between the briber and the o¢ cial, while they are

interested in investigating the willingness to punish corruption where the subject who might

punish does not obtain any economic bene�t from doing so.

2 Experimental design

Using a typical bribery experiment, I consider two participants: a �rm (F hereafter) and a

public o¢ cial (PO hereafter). The design of the experiment tries to capture the characteristics

of corrupt behavior � the reciprocity between the briber and the bribee, the negative externality

over the public, and the probability of being discovered. Focus is placed on the manipulation of a

public o¢ cial�s decisions through the use of a bribe. F is allowed to send some amount of money7The di¤erence between neutral and loaded instructions in this experiment is explained in the following section.

8

to PO, who has to decide between one alternative that is slightly better for himself/herself and

another alternative that is much more favorable to F, but has negative externalities over all the

other participants of the experiment.89

The experiment was designed in completely neutral terms following the design used by Ab-

bink et al. (2002). As explained above, Abbink and Hennig-Schmidt (2002) replicated the

experiment by Abbink et al. (2002) using loaded instead of neutral instructions, �nding no sig-

ni�cant di¤erence in subjects�behavior between the two cases. In the loaded case, subjects were

told that a �rm wanted to run an industrial plant which would have negative consequences for

the public and the public o¢ cial had to decide whether to give permission or not. Prior to the

public o¢ cial�s decision, the �rm could make a private payment to the o¢ cial. In the case that

the public o¢ cial accepted the transfer, there was a certain probability of being discovered and

penalized. In this situation, it was clear that the private payment was an attempt to manipulate

the public o¢ cial�s decision in order to convince him/her to choose the alternative that was

most favorable to the �rm, thereby causing negative externalities over the public. Given that

they did not �nd signi�cant di¤erences in behavior between the neutral and loaded treatments,

it can be deduced that subjects view the situation as being corrupt even when using neutral

instructions. Given this result, and the tradition of using neutral instructions in experimental

economics, I follow the neutral design of Abbink et al. (2002).

The experiment was conducted at the Universitat Autònoma de Barcelona with undergrad-

uate students enrolled in di¤erent majors, who were recruited by public advertisements posted

throughout the campus. The experiment was programmed and conducted with z-Tree software

(Fischbacher, 2007).

The experiment consisted of 20 rounds10, each of which has 4 stages. In the �rst two stages

F has to decide how many tokens to transfer to PO. PO then has to decide whether to accept

the transfer or not (stage 3), and choose between two alternatives (stage 4). The subjects

are endowed with 40 tokens.11 The experiment has four treatments: ff , mm, fm, and mf ,

8As explained below, every time the most favorable alternative to F is chosen, both members of all the other

pairs in the lab are penalized.9See instructions in Appendix 1.10As mentioned above, I am interested in studying corrupt behavior in a long-run relationship between the

briber and the o¢ cial. Repeated interaction is important for the creation of relationships and bonding in bribery

and corruption settings. Due to time constraints, I use 20 rounds instead of the 30 rounds used in Abbink et al.

(2002) .11The reason for the equality of initial endowments is to allow equal payo¤s if non-corrupt behavior is carried

9

where the �rst letter denotes the gender (f-female, m-male) of F and the second letter denotes

the gender of PO. The pairs (F and PO) are anonymously matched and remain unchanged

throughout the experiment.12 In the mixed sessions before entering the lab, subjects are told

that one gender will sit in one part of the lab and the other gender in the other part. When

the instructions are read aloud, they are told that one gender will play one role and the other

gender will play the other role. The objective is not only to check whether men and women

behave di¤erently or not, but also whether men and women behave di¤erently depending on

their partner�s gender.13

Stage 1 F has to decide whether to o¤er a transfer (a bribe) to PO or not. If F decides

to o¤er a bribe, the experiment moves to stage 2. If F decides not to o¤er a bribe, s/he is

asked the following question: What do you think is the probability that your partner will choose

alternative B?14, and the experiment moves to stage 4.

Stage 2 F has to decide how many tokens t to o¤er as a bribe. If F o¤ers any positive amount,

s/he has to pay a �xed transfer cost of 2 tokens.15 For the sake of simplicity, t is de�ned in

integer numbers. They are small enough to ensure that F does not end up with a negative

payo¤. Therefore I take t 2 f1; 2; :::; 10g: After F decides how many tokens to o¤er, s/he is

asked: 1) What do you think is the probability that your partner will accept the tokens you

have o¤ered?; 2) If your partner accepts the o¤er, What do you think is the probability that

s/he will choose alternative B?; and 3) If your partner does not accept the o¤er, What do you

think is the probability that s/he will choose alternative C? 1617 Then, the experiment moves

to stage 3.

out. See Table 3.12A long term relationship between the �rm and the public o¢ cial is represented.13 It would also have been possible to always conduct mixed sessions in which subjects were told the gender

of their partner or even their partner�s name. Telling them their partner�s gender would have placed too much

emphasis on gender, while choosing the second option would have placed the anonymity of the subjects at risk.14Alternative B is the "corrupt" alternative. It is much more favorable to F, but has negative consequences for

the members of all the other pairs in the lab.15The transfer cost represents the cost F has to pay to approach PO. This cost is independent from the fact of

whether PO accepts the transfer or not.16Alternative C implies a costly punishment to F.17The possible answers for the probabilities are: 0, 0.10 - 0.19, 0.20 - 0.29, 0.30 - 0.39, 0.40 - 0.49, 0.50 - 0.59,

0.60 - 0.69, 0.70 - 0.79, 0.80 - 0.89, 0.90 - 0.99, 1.

10

Stage 3 PO has to decide whether or not to accept the bribe o¤ered by F. If PO accepts the

bribe, s/he receives 3t 18, and an integer (n) between 0 and 999 is randomly chosen. If n < 3,

the pair is disquali�ed from the experiment � and are only paid the show-up fee � and if n > 3,the experiment moves to stage 4.19 If PO decides to reject the bribe, then the experiment moves

to stage 4.20

Stage 4 PO has to decide between di¤erent alternatives.

i) If PO has accepted the bribe, s/he has to decide between alternative A and B. Alternative

A is the "non-corrupt" alternative, which is slightly better for PO and has no consequences for

the other participants. Alternative B is the "corrupt" alternative, which is much more favorable

to F but has negative externalities. When alternative B is chosen, 3 tokens are discounted from

the earnings of all the other subjects in the lab. This represents the negative externality that

corruption has over the public.21 The minimum possible deduction is 0 if no pair (apart from

his own) chooses alternative B, and 3 � (g� 1) if all other pairs choose alternative B, where g isthe number of pairs in the lab.22 The payo¤s are de�ned in Table 1.23

Amount transferredAlternative chosen A B A B A B A B A B A B A B A B A B A BFirm 47 67 46 66 45 65 44 64 43 63 42 62 41 61 40 60 39 59 38 58Public official 53 48 56 51 59 54 62 57 65 60 68 63 71 66 74 69 77 72 80 75

1 2 3 4 9 105 6 7 8

Table 1: Payo¤s if the public o¢ cial accepts the bribe

PO�s payo¤s are higher if s/he chooses alternative A rather than B. This is so to re�ect the

18The number of tokens is tripled to show the di¤erence in marginal utility between the �rm (the briber) and

the public o¢ cial. It is assumed that the income of a public o¢ cial is lower than the income obtained in a private

business such as a �rm.19 In the real world, the probability of being discovered is small but the punishment is generally severe. The

design tries to captures this fact.20Abbink et al. (2002) also choose a random number to determine whether or not to disqualify the pair.

They �nd that the inclusion of a small probability of being disquali�ed (0.3%) signi�cantly decreases the level

of corruption in comparison to the case with no probability of being disquali�ed. Therefore, although using a

random number generator is a concern since subjects may not believe that this is a possible outcome, their result

shows that subjects really perceive the disquali�cation as a possible result.21As reported in many papers, corruption is positively related to crime and negatively related to economic

development, among other negative consequences.22There were 13 pairs in the room for 3 sessions , while there were 12 pairs in the remaining session. This

di¤erence is due to technical problems in the laboratory.23The deductions are not included in the following tables.

11

fact that when choosing a corrupt alternative, PO will have to pay some costs, for example,

hiding some information from his/her superiors.

ii) If PO rejects the bribe, s/he has to decide between alternative A and C. Alternative C

implies a costly punishment to F. When PO rejects the bribe, the probability of being disquali�ed

is zero. For this reason, I introduced the possibility that PO may punish F to induce F to evaluate

the corruptibility of PO before o¤ering a bribe. In this situation, even if PO rejects the bribe

and thus the probability of being disquali�ed and penalized is zero, F can be punished by PO.

The payo¤s are shown in Table 2.

Alternative chosen A CFirm 48 36Public official 50 48

Table 2: Payo¤s if the public o¢ cial rejects the bribe

iii) If F does not o¤er a bribe, then PO has to decide between alternative A and B. The

payo¤s are de�ned in Table 3. Again, when the corrupt alternative (B) is chosen, 3 tokens are

discounted from the earnings of all the other subjects in the lab.

Alternative chosen A BFirm 50 70Public official 50 45

Table 3: Payo¤s if the �rm does not o¤er a bribe

When they are caught, both subjects are excluded from the experiment and only paid the

show-up fee. In this case the subjects are asked to complete a questionnaire. To persuade the

subjects to remain in the lab, they are o¤ered an extra payment of 2 euros when they learn that

they are disquali�ed.

The game tree is shown in Figure 1.24

24The letters inside the circles show who is taking the decision, F, PO or N (nature). �3::: shows the amountdeducted from all the other participants each time alternative B is chosen.

12

Figure 1: Game tree

At the end of each round, subjects are informed only about the payo¤ they obtain as a result

of the decisions made by their own group, i.e. they are not informed about the deductions due

to other pairs choosing alternative B until the end of the experiment. In this way, independence

between pairs is maintained.

The �nal payo¤s are calculated as the sum of all the period payo¤s converted into euros �

the rate of conversion is 1.5 euros for 100 tokens plus the show-up fee (3 euros).

After the 20 rounds are over, the subjects are given a questionnaire to rate the likelihood

of their engaging in 16 risky activities on a 1 (very unlikely) to 5 (very likely) scale.25 The

sum of the answers gives the degree of risk loving of the subjects � the higher the sum, the

higher the degree of risk loving. Following Datta Gupta et al. (2005), I decided not to use

lottery choices to measure risk aversion because the experimental task could have in�uenced the

choices of risky lotteries. On the other hand, the in�uence could have gone from risk elicitation

to the experimental task if I had conducted the risk elicitation procedure using lotteries before

the experiment. Moreover, psychometric tests have greater test-retest reliability compared to

lottery choices (Eckel, 2005).

25The questionnaire is the same one used by Datta Gupta et al. (2005), which is a modi�ed version of the

psychometric test in Weber et al. (2002).

13

3 Results

The level of corruption is observed in the frequency and amount of bribes o¤ered and the

frequency of B (corrupt) choices induced by bribes. Table 4 shows the number of subjects and

some preliminary results. No pair was disquali�ed in any session.

Session Number ofsubjects

Averageearnings

Frequency ofbribes offered

Average bribeoffered

Frequency ofacceptance

Frequency ofB choices

Female F ­ Male PO 26 17.03 0.17 3.61 0.80 0.49Male F ­ Female PO 24 16.76 0.35 4.43 0.93 0.41Male F ­ Male PO 26 16.12 0.26 5.96 0.88 0.78Female F ­ Female PO 26 17.72 0.14 3.11 0.49 0.28Total 102 16.91 0.23 4.50 0.82 0.53The average earnings include the show­up fee (3 euros).The average bribe offered is conditional on being positive.The frequency of B choices is conditional on having accepted the bribe.

Table 4: Average values of the main results

The �rst preliminary conclusion that can be drawn from Table 4 is that the frequency of

bribes and the average bribe o¤ered are higher when F is a man. Bribes are more frequently

accepted when they come from a male F, while the lowest rate of acceptance is observed when a

female PO is paired with a female F. The highest frequency of B choices is observed when only

men are playing, while the lowest frequency is observed when only women are playing.

There are six possible scenarios that could take place in each stage game: a bribe is not

o¤ered and alternative A or B is chosen, a bribe is o¤ered and accepted by PO and alternative

A or B is chosen, or a bribe is o¤ered and rejected and alternative A or C is chosen. Given that

the last case (bribe is rejected and alternative C is chosen) is only observed 7 times, the last

two alternatives are considered jointly. Table A5 in Appendix 2 shows how frequently each of

these possible scenarios is observed for each group in each of the sessions. 4 groups (17 times

in total) in the fm session, 6 groups in the mf (42 times) and in the mm session (46 times),

and 2 groups (5 times) in the ff sessions follow the path: F o¤ers a bribe, PO accepts it and

chooses alternative B, i.e. the scenario that is de�ned as corrupt.

The rest of the section is organized as follows. In subsection 3.1, the decisions taken by F

are analyzed. In subsection 3.2, I present the predictions made by F about PO�s behavior, while

in the following two subsections I analyze the decisions taken by PO and the subjects�earnings,

respectively.

14

3.1 Decisions taken by the �rm

The �rst decision F has to take is whether to o¤er a transfer (bribe) to PO. If F decides to o¤er

a transfer, then s/he has to decide how many tokens to o¤er.

The percentage of men that decide to o¤er a bribe at least once is 80%, while the percentage

of women is 65% (p-value=0.247).26 Moreover, the average number of times women o¤er a

bribe (3:2) is smaller (p-value=0.000) than the average number of times men o¤er one (6). On

average, men o¤er a bribe to a male PO 5:2 times, while the number of times they o¤er a bribe

to a female PO is 7 (p-value = 0.025). On the contrary, the number of times women o¤er a

bribe does not depend on their partner�s gender. It is 3:4 when playing with a male PO and 2:8

when their partner was another woman (p-value = 0.398).

RESULT 1: On average, women o¤er a bribe less frequently than men.

Analyzing the second decision � the amount o¤ered once the subject has decided to o¤er a

bribe � I �nd that the average bribe o¤ered by men is 5:11 tokens, and by women, 3:38 tokens

(p-value=0.000). As reported in Table 4, men o¤er 5:96 tokens on average to male POs and 4:43

tokens to female POs (p-value = 0.001), while women o¤er 3:61 and 3:11 tokens, respectively

(p-value = 0.209) As before, men are in�uenced by their partner�s gender, while women are

not. The average bribe o¤ered in Abbink et al. (2002) is 5:68 tokens, which is higher than

4:50, the average in the present experiment.27 Figure 2 shows the histogram of bribes o¤ered,

conditional on being positive. For low amounts, the frequency of women o¤ering a bribe is

clearly higher than the frequency of men. The peak in 6 tokens may be explained by the fact

that by transferring this amount and PO choosing alternative B, both players would get almost

the same payo¤ (62 and 63 tokens).28

26The p-values shown always correspond to the Mann-Whitney U-test.27The average bribe in Abbink et al. (2002) is even higher than the average bribe o¤ered by male �rms in my

experiment. Part of the di¤erence may be due to the di¤erent gender composition of the sample.28This could be the reason why an inequity averse F would decide to o¤er 6 tokens as a bribe. Abbink et al.

(2002) �nd the same peak.

15

0.00

0.05

0.10

0.15

0.20

0.25

0.30

1 2 3 4 5 6 7 8 9 10

Bribe offered

Freq

uenc

y

Male Firm Female Firm

Figure 2: Histogram of bribes o¤ered

RESULT 2: The average transfer o¤ered by women is lower than the average transfer o¤ered

by men.

The �gures below show the frequency of bribes o¤ered and the average bribe � conditional

on being positive � by period. Figure 3 shows that men o¤er a bribe more frequently than

women. Figure 4 shows that � with the exception of periods 2 and 11� the bribe o¤ered by

men is higher than the bribe o¤ered by women, as observed with the aggregated values.29

0.0

0.1

0.2

0.3

0.4

0.5

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Period

Freq

uenc

y of

 bri

bes o

ffere

d

Male Firm Female Firm

Figure 3: Frequency of bribes o¤ered by period

29The econometric analysis below justi�es the aggregation of the data from sessions fm and ff on the one

hand, and mm and mf on the other.

16

0

1

2

3

4

5

6

7

8

9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Period

Ave

rage

 bri

be o

ffere

d

Male Firm Female Firm

Figure 4: Average bribe o¤ered by period

In this experiment, subjects play together for 20 periods. Therefore, not only PO can

reciprocate F by choosing alternative B, but reciprocate PO�s previous behavior by o¤ering

more tokens after s/he has chosen alternative B. This is indeed observed in the experiment. The

average transfer after PO has chosen alternative A is 3:5230, but is 5:64 tokens after s/he has

chosen alternative B.31

In what follows, I econometrically analyze the decisions F has to take, i.e. whether to o¤er

a bribe to PO, and if so, how many tokens to o¤er. Regarding the decision to send a transfer,

I estimate a probit model with a clustered standard error, whose results are shown in Table 5.

The second decision for those who decided to o¤er a bribe is to determine how many tokens to

o¤er. An OLS model is estimated and the results are shown in Table 6. I review the results

by analyzing the e¤ects of the (signi�cant) explanatory variables in both decisions at the same

time.

The independent variables are the gender of the players (the dummy variables take a value of

1 if the player is a woman), the actions PO has taken in the two previous periods, the degree of

risk loving (only in the �rst decision), and the period. I also include interaction terms between

30The average bribe o¤ered when PO has chosen alternative A in the previous period when no bribe was o¤ered

is 3:63, while the average bribe is 3:08 if PO has chosen alternative A in the previous period after accepting a

bribe.31The average bribe o¤ered when PO has chosen alternative B in the previous period when no bribe was o¤ered

by F is 5:86, while the average bribe o¤ered is 5:55 if PO has chosen alternative B in the previous period after

accepting a bribe.

17

the gender of F and the following variables: PO´s gender, PO�s actions and F�s risk loving

degree; the last of which is only in the �rst estimation.32 Degree of risk loving is included as

an explanatory variable in the �rst decision because deciding to o¤er a transfer to PO implies

a risk, since there is a 0.003 probability that the two subjects will be disquali�ed from the

experiment33 and only earn the show-up fee.34 As explained above, the degree of risk loving is

measured in the post experimental questionnaire using the Datta Gupta el al. (2005) version of

the psychometric test in Weber et al. (2002). The variables re�ecting PO�s decisions in previous

periods are included because subjects play a repeated game where they can be in�uenced by their

partner�s previous actions.35 As explained at the beginning of Section 3, there are six possible

scenarios that could take place at each stage of the game, but I consider only 5 possible cases (as

shown in Table A5). Therefore, there are 5 dummies that re�ect PO´s previous actions. The

dummy that takes a value of 1 if no bribe is o¤ered and alternative B is chosen is the excluded

scenario (in the previous period and two periods ago).

32Although the coe¢ cients of these interacted variables are not reported in Tables 5 and 6, they are taken into

account when calculating the marginal e¤ects shown.33 If PO accepts the bribe.34 It is not included in the second estimation because the amount o¤ered as a bribe does not in�uence the

probability of being disquali�ed. Nonetheless, I tested a model including this variable and it does not change any

of the results reported.35 Including more than 2 lags a¤ects the signi�cance of lag 1 and lag 2 due to high levels of correlation and does

not increase the explanatory power of the model.

18

Dependent variable: 1 if F sends a transfer                                     0 otherwise (I) (II) (III) (IV) (V)Female player 1 ­0.1457 ­0.0124 0.0066 0.0011 0.0018

0.06 0.78 0.86 0.98 0.96Female player 2 0.0215 0.0234 0.0243 0.0269 0.0309

0.78 0.62 0.46 0.42 0.37No bribe & Alternative A _ Previous period ­0.4372 ­0.3199 ­0.3101 ­0.3126

0.00 0.00 0.00 0.00Bribe rejected_Previous period 0.1896 0.0665 0.0696 0.0613

0.11 0.49 0.46 0.51Bribe accepted & Alternative A_Previous period 0.1509 0.0243 0.0178 0.0073

0.09 0.69 0.77 0.90Bribe accepted & Alternative  B_Previous period 0.5511 0.5051 0.4957 0.4917

0.00 0.00 0.00 0.00No bribe & Alternative A _ 2 periods before ­0.0534 ­0.0496 ­0.0453

0.39 0.42 0.45Bribe rejected_2 periods before 0.1598 0.1627 0.1523

0.08 0.07 0.07Bribe accepted & Alternative A_2 periods before 0.2870 0.2691 0.2457

0.00 0.00 0.00Bribe accepted & Alternative B_2 periods before 0.3171 0.3040 0.3053

0.00 0.00 0.00Risk loving 0.0036 0.0038

0.26 0.25Period ­0.0073

0.00Number of observations 1020 969 918 918 918Prob > chi2 0.2510 0.0000 0.0000 0.0000 0.0000The table shows the marginal effects evaluated at the average of the independent variables.A constant and interaction terms between the two genders, between Female player 1 and the options,and between Female player 1 and Risk loving are included in the estimations.

Coefficient p­value

Table 5: Probit regression of the decision to o¤er a bribe

Coefficient p­value(I) (II) (III) (IV)

Female player 1 ­1.8090 ­1.5771 ­1.5046 ­1.47780.02 0.02 0.01 0.01

Female player 2 ­1.1701 ­1.0013 ­0.4251 ­0.38210.14 0.13 0.49 0.52

No bribe & Alternative A _ Previous period ­1.9667 ­1.7458 ­1.65530.04 0.01 0.02

Bribe rejected_Previous period ­0.1296 ­1.0124 ­1.03190.89 0.26 0.22

Bribe accepted & Alternative A_Previous period ­2.3819 ­2.9402 ­2.88920.03 0.00 0.00

Bribe accepted & Alternative  B_Previous period ­0.4743 ­1.2868 ­1.27700.56 0.05 0.06

No bribe & Alternative A _ 2 periods before ­0.7439 ­0.75100.23 0.21

Bribe rejected_2 periods before 0.6616 0.66900.44 0.43

Bribe accepted & Alternative A_2 periods before ­0.9105 ­0.91720.14 0.13

Bribe accepted & Alternative B_2 periods before 1.1391 1.10130.02 0.03

Period 0.03260.39

Number of observations 232 214 197 197Prob > F 0.0021 0.0000 0.0000 0.0000A constant and interaction terms between the two genders and between Female player 1and the options are included in the estimations

Table 6: OLS regression of the bribe o¤ered

19

The �rst conclusion that can be derived from Table 5 is that the gender of the

players does not have a signi�cant e¤ect on the decision to o¤er a bribe. The probability of

o¤ering a bribe is not in�uenced by either F�s gender nor by PO�s gender.36 Although result 1

states that male F o¤ers a transfer more frequently than female F, gender is not a signi�cant

explanatory variable when control variables are included. On the other hand, gender has an

e¤ect on the bribe o¤ered. In line with result 2, the bribe o¤ered is lower if F is a woman.

These results show that despite the fact that the probability of o¤ering a bribe is not in�uenced

by F�s gender, the amount o¤ered is in�uenced: female F o¤ers lower bribes than male F. If it

holds that the lower the amount o¤ered as a bribe, the lower the probability that PO will choose

alternative B, then we should observe a lower level of corruption when F is a woman.

The actions PO has taken in previous periods a¤ect the probability of o¤ering a bribe and

the amount o¤ered, which decrease if PO has chosen alternative A after no bribe was o¤ered

by F in the previous period. A possible interpretation is that choosing alternative A in this

situation is seen by F as a sign that no cooperation can be expected from PO. On the other

hand, the probability of o¤ering a bribe increases if PO has accepted a bribe and has chosen

alternative B one or two periods ago. In this case, a corrupt relationship between F and PO is

observed. On the other hand, this dummy variable has a negative sign in the OLS estimation

when it refers to an action taken one period ago, but has a positive sign if the action was taken

two periods ago. This means that the bribe o¤ered is higher if PO has accepted a bribe and has

chosen alternative B two periods ago, but the bribe is lower if this happened one period ago.

A feasible explanation for the negative sign is that F is trying to establish a "cheaper" corrupt

relation. F knows that PO is willing to cooperate and so now tries to achieve cooperation with a

lower bribe. The positive sign of the variable lagged 2 periods shows a persistence of the corrupt

relationship in accordance with the sign observed in Table 4.

Although they do not a¤ect the amount o¤ered, other variables that positively a¤ect the

probability of o¤ering a bribe are those which re�ect the following actions taken by PO two

periods ago: 1) s/he has rejected the bribe, and 2) s/he has accepted the bribe and has chosen

alternative A. In these cases, it could be interpreted that F thinks that PO�s action is due to

a low bribe and not to the fact that PO is unwilling to engage in a corrupt relation (in case

of rejection) or unwilling to choose an alternative that has negative externalities (in case of

accepting the bribe but choosing alternative A). S/he therefore o¤ers a bribe again.

36The fact that PO�s gender is not a signi�cant variable justi�es the aggregation of data in Figures 3 and 4.

20

The variable period has a signi�cant negative e¤ect on the probability of o¤ering a bribe �

the probability decreases across periods � but it is not signi�cant in the OLS estimation.

One variable that is signi�cant in the OLS, but not in the probit estimation, is the dummy

that shows that PO has accepted a bribe and has chosen alternative A in the previous period.

To some extent, its negative sign contradicts the positive e¤ect the variable has in the OLS

estimation when it refers to this action taken two periods ago. If this variable has a positive

sign, it would mean that F thought that PO chose alternative A because the bribe was not high

enough. However, its negative sign means that F is replying to PO�s "nasty" action (he has not

chosen the alternative most favorable to F) with another "nasty" action � lowering the bribe

o¤ered.

Finally, contrary to what could be expected, the risk loving degree does not have a signi�cant

e¤ect on the decision to o¤er a bribe.

The main conclusion from the previous econometric analysis is the same as result 2.

3.2 Beliefs regarding the public o¢ cial�s actions

After F has made his/her decisions, F is asked about his/her beliefs regarding the actions PO

would take. In case F has not o¤ered a bribe to PO, s/he is asked: What do you think is

the probability that your partner will choose alternative B?37 If F has o¤ered a bribe, s/he is

asked about the probability of PO accepting the transfer and the probability of PO choosing

alternative B if PO accepts the transfer, or alternative C if PO rejects the transfer.

When F does not o¤er a bribe, men assign on average a 0:18 probability of PO choosing

alternative B, while women assign on average a probability of 0:10 (p-value=0.000). Women are

not in�uenced by their partner�s gender, while men are. Men assign a higher probability to a

male PO (0:23) than to a female PO (0:13) (p-value=0.017).

When F o¤ers a bribe, men assign a higher probability to PO accepting the bribe (0:81) and

PO choosing alternative B (0:60) than women (0:68 and 0:36, respectively).38

RESULT 3: Men assign a higher probability to PO accepting the bribe and choosing the

corrupt alternative than women.

37Recall that alternative B is the "corrupt" alternative.38Both comparisons between female and male probabilities yield p-values = 0.000.

21

The average probability assigned to a male PO accepting the bribe is higher than the prob-

ability assigned to a female PO � 0:80 vs. 0:74 (p-value=0.084). The same is observed with

respect to the probability of PO choosing alternative B � 0:59 vs. 0:45 (p-value=0.001). Male

Fs do not assign di¤erent probabilities to PO accepting the transfer according to their partner�s

gender � 0:81 vs. 0:82 (p-value = 0.812) � while women do di¤erentiate � 0:78 vs. 0:56

(p-value = 0.000). Regarding the probability of choosing the alternative that is most favorable

to F, men and women assign di¤erent probabilities to male and female POs. The probability

men assign to PO choosing alternative B is 0:70 when they are playing with men, and 0:53

when they are playing with women (p-value = 0.000). The probabilities assigned by women are

0:42 and 0:29, respectively (p-value = 0.082). When F is asked about the probability of PO

choosing alternative C in case PO rejects the o¤er, women assign a higher probability than men

� 0:40 vs. 0:25 (p-value = 0.000). None of them assigns di¤erent probabilities according to

PO�s gender.

The di¤erence in expectations about the behavior of male and female public o¢ cials shows

that Fs expect women to be less corrupt than men.

RESULT 4: Male and female �rms expect female public o¢ cials to choose the corrupt alter-

native less frequently than male o¢ cials.

Figures 5 and 6 show the average di¤erence between the assigned probabilities and the real

frequencies. Figure 5 shows that women tend to overestimate the frequency, while men tend to

underestimate it. On the other hand, Figure 6 shows no signi�cant di¤erences between women

and men in terms of the accuracy of assigning probabilities.39

39The lack of signi�cance of player 2�s gender in the econometrical analysis below justi�es the aggregation of

sessions fm with ff , and mm with mf .

22

­0.40

­0.30

­0.20

­0.10

0.00

0.10

0.20

0.30

0.40

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Period

Prob

abili

ty ­ 

Freq

uenc

y

Male Firm Female Firm

Figure 5: Probability of accepting the bribe minus frequency of acceptance

­0.4

­0.3

­0.2

­0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Period

Prob

abili

ty ­ 

Freq

uenc

y

Male Firm Female Firm

Figure 6: Probability of choosing alternative B minus frequency of B choices

Table 7 shows two tobit estimations for the probabilities of accepting the bribe and choosing

alternative B when F decides to o¤er a bribe.40 The estimations show that once control variables

are considered, the gender of the players is no longer signi�cant. The amount o¤ered as a bribe

has a positive and signi�cant e¤ect on both probabilities. The other variable that a¤ects both

probabilities is the dummy that shows that two periods ago PO has accepted the bribe and has

chosen alternative B. As expected, it has a positive e¤ect as it does in the models estimating

the probability of o¤ering a bribe and the amount o¤ered. The �rst probability also increases

40 I estimate these two models and not the probability of choosing alternative B when F is not o¤ering a bribe

because I want to focus on PO�s decisions when F has o¤ered a bribe.

23

if two periods ago PO has chosen alternative A after no bribe or after a bribe was o¤ered by F.

The last variable also has a positive and signi�cant e¤ect in the model explaining the decision

to o¤er a bribe. Nonetheless, one could expect the previous variable to have a negative sign

instead of a positive one. The degree of risk loving has a positive and signi�cant e¤ect � the

higher the risk loving degree of F, the higher the probability s/he assigns to PO accepting the

bribe.

The probability assigned to PO choosing alternative B decreases if PO has accepted the bribe

and has chosen alternative A in the previous period. This variable also has a signi�cant and

negative e¤ect on the amount o¤ered as a bribe.

Random effects Tobit estimationsDependent variable: Accept Alternative B

Female player 1 ­0.0700 ­0.11930.4 6 0.11

Female player 2 0.0476 ­0.03340.6 2 0.63

Amount offered as a bribe 0.0222 0.03750.0 1 0.00

No bribe & Alternative A _ Previous period ­0.0245 ­0.05010.6 7 0.44

Bribe rejected_Previous period ­0.0895 ­0.06930.2 8 0.46

Bribe accepted & Alternative A_Previous period 0.0197 ­0.13000.7 8 0.10

Bribe accepted & Alternative  B_Previous period 0.0304 0.06980.6 0 0.26

No bribe & Alternative A _ 2 periods before 0.1618 ­0.09260.0 2 0.23

Bribe rejected_2 periods before 0.0567 ­0.01650.5 1 0.86

Bribe accepted & Alternative A_2 periods before 0.1322 ­0.03680.0 8 0.66

Bribe accepted & Alternative B_2 periods before 0.1665 0.15490.0 1 0.04

Risk loving 0.0166 ­0.00170.0 4 0.78

Period ­0.0017 ­0.00220.5 5 0.51

Number of observations 197 197Prob > chi2 0.0125 0.0000The table shows the marginal effects evaluated at the average of the independent variables.A constant and interaction terms between the two genders, between Female player 1 and the options,and between Female player 1 and Risk loving are included in the estimations.

Table 7: Tobit estimations of the probabilities assigned by the �rm when o¤ering a bribe to

the public o¢ cial

24

3.3 Decisions taken by the public o¢ cial

The decisions PO has to take depend on whether F has decided to o¤er a bribe to PO or not.

As the focus of the paper is on the manipulation of a public o¢ cial�s decisions through the use

of a bribe, I analyze the decisions PO takes when F has o¤ered the public o¢ cial a bribe. In

this situation, PO�s �rst decision is whether to accept the bribe or not. Then, as explained in

section 2, PO has to decide between two alternatives.

The percentage of male POs that are o¤ered a bribe at least once in the 20 periods is

73%, while the percentage of women is 72%. When receiving an o¤er, the average frequency of

acceptance is 85% for men and 79% for women (p-value = 0.292).

RESULT 5: The frequencies with which men and women accept a bribe are not statistically

di¤erent.

As Table 4 shows, the frequency with which women accept the bribe is 93% when playing

with a male F and 49% when playing with a female F (p-value = 0.000). The frequency with

which men accept the o¤er when they are playing with a male F is 88%. When F is a woman

it is 80% (p-value = 0.225). In this case, women behave di¤erently if playing with a man or

with another woman, while men do not. Moreover, the di¤erence between men and women�s

frequency of acceptance when they are playing with a male F is not statistically signi�cant (p-

value=0:314). When a male F o¤ers a bribe, it is accepted in 88% of the cases if he is playing

with another man, and in 93% of the cases when the PO is a woman. When F is a woman, the

frequency of acceptance of the bribe o¤ered by her di¤ers (p-value=0.004) if PO is a man (80%)

or a woman (49%).

The next step is to analyze the alternative PO chooses once s/he has accepted the bribe

o¤ered by F. In this situation, men choose alternative B in 67% of the cases, while women do

so in 39% of the cases (p-value = 0.000).

RESULT 6: Women choose alternative B less frequently than men.

Figures 7 and 8 show the frequency of B choices by period and by bribe amount, respectively.

With some exceptions, both �gures show that the frequency is higher for men than for women.

Figure 7 also shows that no male PO takes the action that favors F in period 1, but they do

so in the following periods. This is probably a form of strategic behavior to push F to increase

25

the bribe o¤ered. In period 2 the bribe o¤ered � and accepted � is not higher (it was 3:67;

while in period 1 it was 4:07), but 80% of male POs choose alternative B. The evolution of the

frequency of B choices is rather similar to the evolution of the bribe amount.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20Period

Freq

uenc

y of

 B c

hoic

es

Male Public Official Female Public Official

Figure 7: Frequency of B choices by period

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1 2 3 4 5 6 7 8 9 10Amount of the bribe

Freq

uenc

y of

 B c

hoic

es

Male Public Official Female Public Official

Figure 8: Frequency of B choices by amount of the bribe

The frequency of B choices by the male PO when playing with another man is 78%, but when

playing with a woman the percentage is 49% (p-value = 0.004). The percentages for the female

PO are 41% and 28%, respectively (p-value = 0.300). In this case, it is men and not women

who di¤erentiate according to their partner�s gender, although the opposite was observed with

respect to the frequency of acceptance. These values are shown in Table 4.

26

It is interesting to note that the highest frequency of B choices is observed when only men

are playing (78%), while the lowest frequency is observed when only women are playing (28%).

Figures A1 and A2 in Appendix 2 show the frequency of B choices for the four di¤erent

sessions. Figure A1 shows that the frequency of B choices by male POs di¤er according to their

partner�s gender. On the other hand, in line with previous results, the frequency of female POs

does not di¤er depending on F�s gender. Figure A2 shows that the frequency of females�B

choices is lower than the frequency of males�, independently of F�s gender.

As explained above, F is asked to assign a probability to the event that PO will accept the

bribe and that PO will choose alternative B. It turns out that the average assigned probability

to PO accepting the bribe is very close to the real frequency when only women are playing � the

di¤erence is 0:01� but is not so accurate in the other cases. When only men are playing, they

underestimate both frequencies by 0:07 points on average in the �rst probability and by 0:08

points in the second one. When a female F is playing with a male PO, they underestimate

the frequency of acceptance and B choices by 0:02 and 0:07 points, respectively. The highest

di¤erences are observed when a male F is paired with a female PO. They underestimate the

probability of acceptance by 0:11 points, and the probability of PO choosing alternative B by

0:12 points.

In the remainder of this subsection I will show the results of the econometric analysis of

the two POs�decisions. The results are shown in Tables 8 and 9. The estimated models are

probits with clustered standard errors. In the �rst model, the dependent variable takes a value

of 1 if PO accepts the bribe and 0 otherwise. In the second model, the dependent variable takes

value 1 if PO chooses the alternative that is most favorable to F (alternative B), and takes value

0 if PO chooses alternative A. The independent variables are the gender of both players (the

dummy variables take a value of 1 if the player is a woman); the amount of the bribe o¤ered (and

accepted by PO) in the current period, one period ago, and two periods ago; the degree of risk

loving (only in the �rst decision), and the period. I also include interaction terms between the

gender of the two players, between the gender of PO and the amount of the (current and lagged)

bribes, and between PO�s gender and the risk loving degree. Risk loving degree is included in

the decision of whether to accept the bribe or not because the decision implies a risk. That is, if

PO accepts the bribe, then a number is randomly chosen to decide if the pair will be disquali�ed

or not. If the pair is not disquali�ed, then PO has to choose one alternative. In this decision the

risk loving degree plays no role, given that there is no risk of being disquali�ed at this point.

27

Coefficient p­value(I) (II) (III) (IV) (V) (VI)

Female player 2 ­0.0274 0.0005 0.0152 0.0204 0.0145 0.01600.70 1.00 0.80 0.70 0.78 0.75

Female player 1 ­0.2609 ­0.2173 ­0.2614 ­0.2587 ­0.2381 ­0.23500.01 0.07 0.01 0.01 0.02 0.02

Amount of bribe 0.0216 0.0380 0.0479 0.0490 0.04890.08 0.00 0.00 0.00 0.00

Amount bribe_Previous period ­0.0165 ­0.0221 ­0.0216 ­0.02100.17 0.10 0.10 0.12

Amount bribe_2 periods ago ­0.0051 ­0.0054 ­0.00500.69 0.67 0.68

Risk loving ­0.0016 ­0.00190.75 0.72

Period ­0.00150.79

Number of observations 232 232 214 197 197 197Prob > chi2 0.0007 0.0171 0.0000 0.0000 0.0000 0.0000The table shows the marginal effects evaluated at the average of the independent variables.A constant and interaction terms between the two genders, between Female player 2 and the amount of the bribe,and between Female player 2 and Risk loving are included in the estimations.

Table 8: Probit regression of the decision to accept the bribe

Coefficient p­value(I) (II) (III) (IV) (V)

Female player 2 ­0.3390 ­0.3513 ­0.3462 ­0.3579 ­0.34510.01 0.00 0.00 0.00 0.00

Female player 1 ­0.0518 ­0.1293 ­0.0803 ­0.1038 ­0.09990.42 0.26 0.42 0.34 0.36

Amount of bribe 0.0962 0.1020 0.1151 0.11410.00 0.00 0.00 0.00

Amount bribe_Previous period 0.0178 0.0293 0.02600.24 0.12 0.14

Amount bribe_2 periods ago ­0.0321 ­0.03220.25 0.25

Period 0.07890.47

Number of observations 190 190 176 164 164Prob > chi2 0.0150 0.0000 0.0000 0.0000 0.0000The table shows the marginal effects evaluated at the average of the independent variables.A constant and interaction terms between the two genders and between Female player 2 and the amountof the bribe are included in the estimations.

Table 9: Probit regression of the decision to choose alternative B

Although PO�s gender does not have a signi�cant e¤ect on the decision to accept the bribe

when we calculate the marginal e¤ect on the average of the variable re�ecting F�s gender, it is

signi�cant when we calculate the e¤ect when F is 1, i.e. F is a woman. In this case, the marginal

e¤ect is �0:3563 and its p-value is 0.02. This means that if F is a woman, the probability that afemale PO will accept the bribe is 0:3563 points lower than the probability that a male PO will

accept the same bribe. The gender of PO has a signi�cant and negative e¤ect on the decision

to choose the corrupt alternative. If PO is a woman, the probability of choosing the alternative

most favorable to F is 0:3451 points lower � ceteris paribus� than the probability that a male

PO will choose this alternative. This result is similar in some ways to the result in Dreber and

28

Johannesson (2008). They �nd that men are more prone to lie in order to secure a higher payo¤

to the detriment of their partner�s payo¤ than women, while I �nd that men are more prone to

choose an alternative that negatively a¤ects the other players.

When F is a woman, the probability that PO will accept the bribe decreases by 0:2350 points.

Nonetheless, this variable is not signi�cant in Table 9. But if we calculate the marginal e¤ect

of Female �rm when PO is a man � instead of doing it on the average of the variable � this

variable has a signi�cant and negative e¤ect. The marginal e¤ect is �0:2131 and the p-value is0.03. This result is in line with the previous result that shows that F�s gender is important to

male POs when deciding whether to choose alternative B or not. The probability that a male

PO will choose alternative B is 0:2131 points lower when F is a woman rather than a man.

As expected, the bribe amount o¤ered in the current period has a positive and signi�cant

e¤ect in both decisions. The higher the amount, the higher the probability of accepting the

bribe and choosing alternative B. One more token o¤ered as a bribe increases the probability

of acceptance by 0:0489 points and the probability of choosing alternative B by 0:1141 points.

Contrary to Eckel and Grossman (2001), I do not �nd that females are more likely to accept

lower o¤ers. Nonetheless, what I do �nd is that bribes o¤ered to women are lower than bribes

o¤ered to men (as shown in subsection 3.1). Therefore, �rms might expect women to accept

lower bribes.

The other variables included in the estimations are not signi�cant. The most important

conclusion of this analysis is summarized in the following results.

RESULT 7: If the public o¢ cial is a woman, the probability of accepting the bribe is lower

when the �rm is also a woman.

RESULT 8: If the public o¢ cial is a woman, the probability of choosing the corrupt alterna-

tive is lower.

3.4 Earnings

The average initial earnings41 are 15:76 euros for male F and 15:24 euros for female F (p-value =

0.015). When playing with a male PO, male F earns 15:94 euros on average, while when playing

41The initial earnings are the earnings before subtracting the corresponding tokens due to the number of times

alternative B has been chosen.

29

with a female PO they earn 15:56 euros on average (p-value = 0.650). The corresponding

values for female F are 15:48 and 15:00 euros (p-value = 0.081). The initial earnings of PO are

15:94 and 15:21 for men and women, respectively (p-value = 0.140). The male PO earns 15:87

euros when playing with a man and 15:25 when playing with a woman (p-value = 0.418). The

values for the female PO are 15:25 and 15:18 euros, respectively (p-value = 0.186). Therefore,

before deductions, male �rms earn more than female F on average, and female F earns more

when playing with males than when playing with females. There is no di¤erence in PO�s initial

earnings.

Given the di¤erent number of POs that choose alternative B, we have di¤erent deductions

in each session. The average deduction in the mm session is 2:78 euros, in the mf session it is

2:02 euros and in the fm and ff sessions they are 1:33 and 0:37 euros, respectively.42

Therefore, the �nal earnings after subtracting the above deductions are 13:34 euros for male

F and 14:39 euros for female F (p-values=0,000). Note that the di¤erence in the �nal earnings

has the opposite sign as does the di¤erence in the initial earnings.

Male F earns 13:16 and 13:54 euros when playing with men and women, respectively (p-value

= 0.186). The average earnings of female F are 14:15 and 14:63 euros, respectively (p-values =

0.057). The average �nal earnings of male PO are 13:52 euros, while the earnings of female PO

are 14:36 (p-values = 0.002). The average earnings of male PO are 13:09 and 13:92 (p-value =

0.012), while those of female PO are 13:98 and 14:81 (p-values = 0.004) when playing with a

male F and a female F, respectively. In both cases public o¢ cials earn more when playing with

a female F due to the di¤erence in the deductions. The main results are summarized as follows.

RESULT 9: Before deductions, male �rms earn on average more than female �rms. Con-

sidering the deductions, female �rms earn more than male �rms. Also, female public o¢ cials

earn more than male public o¢ cials. The di¤erence is due to higher deductions when men are

playing.

42Recall that mm means Male player 1 - Male player 2, mf means Male player 1 - Female player 2, fm Female

player 1 - Male player 2, and ff Female player 1 - Female player 2.

30

4 Conclusions

The aim of this paper was to study in a controlled environment whether women and men behave

in di¤erent ways with respect to corruption � as suggested in the papers using �eld data� or,

on the contrary, they behave in a similar way. In the experiment, participants took one of two

roles, that of a �rm or that of a public o¢ cial. The possibility of corruption was introduced

by allowing the �rst player (the �rm) to send some amount of money as a bribe to the second

player (the public o¢ cial) in the hope of persuading the o¢ cial to take a decision favorable to

the �rm, although this decision had negative externalities over all the other participants in the

experiment.

The percentage of male �rms that decided to o¤er a bribe to the public o¢ cer at least once

was 80%, while the percentage of female �rms that did so was 65%. Moreover, the average

number of transfers was 6 for men and 3:1 for women; statistically di¤erent quantities. The

average amounts o¤ered � conditional on being positive � were also statistically di¤erent.

They were 5:11 and 3:38 for male and female �rms, respectively.

The estimations show that even when controlling for the previous actions of the public o¢ cer,

the public o¢ cer�s gender and the �rm�s risk aversion, a lower amount was transferred if the

�rm is a woman. This result implies that if the manipulation of the public o¢ cer�s decisions

positively depends on the amount o¤ered as a bribe � as I found� then the probability of

observing a corrupt relationship will be lower if the �rm is a woman. When estimating the

probabilities of the public o¢ cer�s actions, men assigned a higher probability than women to

the public o¢ cer accepting the bribe and choosing the corrupt alternative. Moreover, both male

and female �rms expected female public o¢ cials to choose the corrupt alternative less frequently

than male o¢ cials.

The average frequencies of bribe acceptance were not statistically di¤erent for male and

female public o¢ cials. When the bribe was accepted, men chose alternative B in 67% of the

cases, while women did so in 39% of the cases; statistically di¤erent percentages. The highest

average percentage of B choices was observed when only men were playing (78%), while the

lowest was observed when only women were playing (28%). The probit models estimated show

that if the public o¢ cial is a woman, the probability of accepting the bribe is lower (if the �rm

is also a woman), as well as the probability of choosing the corrupt alternative.

Given the results mentioned above, the conclusion is in line with the �eld data papers. That

31

is, women are less corrupt than men. Thus, increasing participation by women in the labor force

and politics would be expected to help in �ghting corruption.

One question that arises at this point is whether the observed di¤erences are due to stereo-

typing behavior or are simply inherent di¤erences between males and females. If women are

stereotyped as being less corrupt, they would receive less bribe o¤ers and would therefore be

less frequently involved in corrupt activities. This would not necessarily be due to inherent

gender di¤erences, but to the stereotyped behavior of the bribers.

With the raw numbers, I found that men o¤ered bribes more frequently to females than

to male public o¢ cials, although they o¤ered them lower bribes. With respect to the assigned

probabilities, I found that both genders assigned a lower probability to their partner choosing

alternative B when playing with a female public o¢ cer instead of a male public o¢ cer. This

result might show some stereotyping behavior or a learning process by the �rms, given that

female o¢ cers indeed chose alternative B less frequently than males. Despite these results, the

partner�s gender is generally not signi�cant in the econometric analysis of the decisions and

assigned probabilities. Partner�s gender is only signi�cant in the estimation of the probability of

accepting the bribe, and when the public o¢ cial is a man in the estimation of the probability of

choosing the corrupt alternative. Both probabilities are lower if the �rm is a woman. Therefore,

I observe only weak evidence of stereotyping behavior.

According to Gottfredson and Hirschi (1990), one explanation of gender di¤erence in crime

is a di¤erence in self-control. Self-control is "the extent to which they are vulnerable to the

temptations of the moment" or "the extent to which they are restrained from criminal acts".

This behavior can also be extended to di¤erences in corruption. This would mean that women

exercise greater self-control and therefore refrain from engaging in corrupt behavior. Moreover,

as Gottfredson and Hirschi (1990) point out, people with low self-control tend to be more ego-

centric and disinterested in others�needs. In relation with corruption, one possible explanation

for the di¤erence observed in the experiment is that women are more sensitive to others�losses

and that is why they choose the corrupt alternative with negative externalities over all the other

participants less frequently.

To reach a de�nitive conclusion as to why gender di¤erence is observed, more studies need to

be conducted on this subject. This paper is an attempt to study gender di¤erences in corruption

through a lab experiment, albeit further research should shed more light on this topic.

32

References

[1] Abbink, Klaus and Heike Hennig-Schmidt (2002), "Neutral versus loaded instructions in a

bribery experiment". Working paper, University of Nottingham and University of Bonn.

[2] Abbink, Klaus, Bernd Irlenbusch, and Elke Renner (2002), "An experimental bribery

game". The Journal of Law, Economics and Organization, 18 (2), pp. 428-454.

[3] Alatas, Vivi, Lisa Cameron, Ananish Chaudhuri, Nisvan Erkal, and Lata Gangadharan

(forthcoming), "Gender and Corruption: Insights from an experimental analysis". Southern

Economic Journal.

[4] Chaudhuri, Ananish and Lata Gangadharan (2003), "Gender di¤erences in trust and reci-

procity". Working Paper, Economics, The University of Auckland.

[5] Croson, Rachel and Uri Gneezy (2008), "Gender di¤erences in preferences". Journal of

Economic Literature, forthcoming.

[6] Datta Gupta, Nabanita, Anders Poulsen, and Marie-Claire Villeval (2005), "Male and Fe-

male Competitive Behavior. Experimental Evidence". GATE Working Paper 05/12.

[7] Dreber, Anna and Magnus Johannesson (2008), "Gender di¤erences in deception". Eco-

nomics Letters, 99, pp. 197-199.

[8] Dollar, David, Raymond Fisman, and Roberta Gattia (2001), "Are women really the

�Fairer� sex? Corruption and women in government". Policy research report on gender

and development. Journal of Economic Behavior and Organization, 46(4), pp. 423-429.

[9] Eagly, Alice and Wendy Wood (1999), "The origins of sex di¤erences in human behavior:

Evolved dispositions versus social roles". American Psychologist, 54, pp. 408-423.

[10] Eckel, Catherine and Philip Grossman (2000), "Di¤erences in the economic decision of men

and women: experimental evidence". In Handbook of Experimental Economics Results. Ed

by C Plott and V.L. Smith.

[11] Eckel, Catherine and Philip Grossman (2001), "Chivalry and Solidarity in Ultimatum

Games�. Economic Inquiry, 39, pp. 171-188.

33

[12] Eckel, Catherine and Philip Grossman (2002), �Sex Di¤erences and Statistical Stereotyping

in Attitudes Toward Financial Risk�. Evolution and Human Behavior, 23 (4), pp. 281-295.

[13] Eckel, Catherine and Rick Wilson (2004a), �Whom to Trust? Choice of Partner in a Trust

Game�. Working Paper, Virginia Polytechnic.

[14] Eckel, Catherine and Rick Wilson (2004b), �Conditional Trust: Sex, Race and Facial Ex-

pressions in a Trust Game�. Working Paper, Virginia Polytechnic.

[15] Fischbacher, Urs (2007), "Z-tree. Zurich toolbox for readymade economic experiments".

Experimental Economics, 10(2), pp.171-178.

[16] Frank, Björn and Günther Schulze (2000), "Does economics make citizens corrupt?". Jour-

nal of Economic Behavior and Organization, 43 (1), pp. 101-113.

[17] Gottfredson, Michael and Travis Hirschi (1990), "A general theory of crime". Stanford:

Stanford University Press.

[18] Phyllis, Dininio and Sahr John Kpundeh (1999), "A handbook on �ghting corruption".

Technical Publication Series. Center for Democracy and Governance, U.S. Agency for In-

ternational Development.

[19] Johnson, Johnnie and Philip Powell (1994), �Decision making, risk and gender: Are man-

agers di¤erent?�. British Journal of Management, 5, pp. 123-138.

[20] Schulze, Günther and Björn Frank (2003), "Deterrence versus intrinsic motivation: exper-

imental evidence on the determinants of corruptibility". Economics of Governance, 4 (2),

pp. 143-160.

[21] Schubert, Renate, Matthias Gysler, Martin Brown, and Hans-Wolfgang Brachinger (1999),

"Financial decision-making: Are women really more risk averse?". American Economics

Review Paper and Proceedings, 89, pp. 381-385.

[22] Selten, Reinhard and Axel Ockenfels (1998), "An experimental solidariy game". Journal of

Economic Behavior and Organization, 34, pp. 517-539.

[23] Snijders, Chris and Gideon Keren (2004), �Do you trust? Whom do you trust? When do

you trust?�In S.R. Thye, E.J. Lawler, M.W.Macy, H.A. Walker (Eds.), Advances in group

processes, pp. 129-160. Amsterdam: JAI, Elservier Science.

34

[24] Swamy, Anand, Stephen Knack, Young Lee, and Omar Azfar (2001), "Gender and corrup-

tion". Journal of Development Economics, 64 (1), pp. 25-55.

[25] Torgler, Benno and Neven T. Valev (2006), "Women and illegal activities: gender di¤erences

and women�s willingness to comply over time". Working Paper Nr. 2006-15. Center for

Research in Economics, Management and the Arts (CREMA).

[26] Weber, Elke, Ann-Renée Blais and Nancy Betz (2002), "A domain-speci�c risk attitude

scale: measuring risk perceptions and risk behaviors". Journal of Behavioral Decision Mak-

ing, 15 (4), pp. 263-290.

[27] Wood, Wendy and Alice Eagly (2002), "A cross-cultural analysis of the behavior of women

and men: Implications for the origins of sex di¤erences". Psychological Bulletin, 128, pp.

699-727.

35

5 Appendix 1: Instructions

(Original text in Spanish)

Thank you for participating in this experiment on decision making. You will be paid 3 euros

for showing up plus the money you earn during the experiment, which will depend on your and

other participants�decisions. At the end of the session you will be paid in private.

From now on, you will not be allowed to speak with the other participants. If you have

any questions or doubts at any time during the experiment, please raise your hand and we will

attend to you personally.

There are two types of participants in the experiment: player type 1 (hereafter Player 1) and

player type 2 (hereafter Player 2). Players 1 are those that have an odd number and Players

2 those that have an even number. At the beginning of the experiment, each Player 1 will be

matched with one Player 2 in an anonymous way. This match will be maintained throughout

experiment.

The experiment consists of 20 identical and independent rounds. Each round consists of 4

stages which will be explained in short. At the beginning of each round, each participant will be

assigned 40 tokens, meaning that each participant will have 40 tokens to be used in this round.

Stage 1

Player 1 has to decide whether to send a transfer to Player 2 or not.

� If he does, his credit is reduced by 2 tokens (which is a �xed cost for transferring tokens)and the experiment moves to stage 2.

� If Player 1 decides not to send a transfer to Player 2, his credit remains unchanged andthe experiment moves to stage 4.

Stage 2

Player 1 decides how many tokens to transfer to Player 2 from among 1, 2, 3, 4, 5, 6, 7, 8,

9, or 10 tokens. The experiment moves to stage 3.

Stage 3

Player 2 decides whether to accept the transfer or not.

36

� If Player 2 decides to accept the transfer, then Player 1�s credit is reduced by the amounttransferred and Player 2�s credit is increased by the triple of the amount transferred by

Player 1. For example, if Player 1 sends 2 tokens, Player 2 receives 6 tokens. After this,

an integer ranging from 0 to 999 is randomly drawn by the computer (every number has

the same probability of being chosen).

� If the randomly drawn number is 0, 1, or 2, then the pair is disquali�ed. This means

that the experiment ends for these two players and their earnings in the experiment

are 0. They will only receive 3 euros for showing up. These players have to remain

in their seats in silence and �ll in a questionnaire that will be distributed to them.

For the other participants, the experiment will continue normally.

� If the randomly drawn number is 3, 4, 5, ...or 999, then the experiment moves to

stage 4.

� If Player 2 decides not to accept the transfer o¤ered by Player 1, then the credit remainsunchanged (the transfer cost from stage 1, however, is paid by Player 1). The experiment

moves to stage 4.

Stage 4

Player 2 chooses one alternative depending on what has happened until that time, i.e.

whether Player 1 has transferred tokens and in case he did, whether Player 2 has accepted

the transfer or not.

1. If Player 1 decided not to o¤er a transfer, Player 2 has to choose between alternatives A

and B.

2. If Player 1 o¤ered a transfer that was accepted by Player 2, then Player 2 has to choose

between alternatives A and B.

3. If Player 1 o¤ered a transfer that was rejected by Player 2, then Player 2 has to choose

between alternatives A and C.

This is shown in the following table.

37

AlternativesStage 1 Stage 2 Stage 3 Stage 4

Player 1 Player 1 Player 2 Player 2Decides not to send tokens ­­­­­­­ ­­­­­­­ A or BDecides to send tokens Decides the amount of tokens Accepts the tokens A or BDecides to send tokens Decides the amount of tokens Does not accept the tokens A or C

Situation

Table A1

Apart from the corresponding payo¤s in each situation (see tables), for every pair that chooses

Alternative B (i.e. for each Player 2 that does so) the payo¤ of all the other participants will be

decreased by 3 tokens. These 3 tokens are not included in the tables.

CASE 1: Player 1 decided not to o¤er a transfer to Player 2.

The payo¤s are as follows.

Alternative A BFirm 50 70Public official 50 45

Table A2

This means that if Player 2 chooses Alternative A, the payo¤s are 50 tokens for both players.

If, on the contrary, Alternative B is chosen, the payo¤s are 70 tokens for Player 1 and 45 for

Player 2. If Alternative B is chosen, 3 tokens will be deducted from the payo¤s of both members

of all the other pairs in the room. The amounts in the table are the total payo¤s corresponding

to this round, including the initial tokens (40 and the added tokens for each alternative chosen.

CASE 2 : Player 1 decided to o¤er a transfer of tokens, which was accepted by Player 2.

In this case the payo¤s depend on the amount transferred. The amounts in the following

table are the total payo¤s corresponding to this round of the experiment, including the initial

tokens (40), the added tokens for each alternative chosen, the transfer made, and the cost of

making the transfer.

Transfer amountAlternative A B A B A B A B A B A B A B A B A B A BFirm 47 67 46 66 45 65 44 64 43 63 42 62 41 61 40 60 39 59 38 58Public official 53 48 56 51 59 54 62 57 65 60 68 63 71 66 74 69 77 72 80 75

1 2 3 4 9 105 6 7 8

Table A3

For example,

38

� If 2 tokens are transferred and Alternative A is chosen, the payo¤s are 46 tokens for Player1 and 56 tokens for Player 2. If Alternative B is chosen, then the payo¤s are 66 and 51

tokens, respectively. If Alternative B is chosen, 3 tokens will be deducted from the payo¤s

of each member of all the other pairs in the room.

� If 5 tokens are transferred and Alternative A is chosen, the payo¤s are 43 tokens for Player1 and 65 tokens for Player 2. If Alternative B is chosen, then the payo¤s are 63 and 60

tokens, respectively. If Alternative B is chosen, 3 tokens will be deducted from the payo¤s

of each member of all the other pairs in the room.

� If 10 tokens are transferred and Alternative A is chosen, the payo¤s are 38 tokens for

Player 1 and 80 tokens for Player 2. If Alternative B is chosen, then the payo¤s are 58

and 75 tokens, respectively. If Alternative B is chosen, 3 tokens will be deducted from the

payo¤s of each member of all the other pairs in the room.

CASE 3 : Player 1 o¤ered a transfer and it was rejected by Player 2.

In this case the payo¤s are as follows.

Alternative A CFirm 48 36Public official 50 48

Table A4

This means that if Player 2 chooses Alternative A, the payo¤s are 48 tokens for Player 1 and

50 tokens for Player 2. If Alternative C is chosen, the payo¤s are 36 tokens for Player 1 and 48

tokens for Player 2. The above amounts are the total payo¤s corresponding to this round of the

experiment, including the initial tokens (40), the tokens added or deducted for each alternative

chosen, and the cost of making a transfer.

DEDUCTIONS

From the previous payo¤s we should deduct an amount depending on how many type 2

players have chosen Alternative B. For each Player 2 of the other pairs that choose Alternative

B, we have to deduct 3 tokens from the payo¤s. For example, if there are 5 pairs (excluding

yours) that have chosen Alternative B, then we have to deduct 15 tokens from your and your

partner�s payo¤s (5 pairs * 3 tokens).

39

The minimum deduction is 0 (if none of the other pairs chooses Alternative B). The maximum

deduction is 3 times the number of pairs that are now in the room (excluding yours). For

example, if there are 13 pairs in the room, it means that there are 12 pairs apart from your own.

Hence, the maximum possible deduction is 3 * 12 = 36 tokens.

After stage 4, the round ends. The total earnings of the experiment are the sum of the payo¤s

obtained in the 20 rounds converted into euros plus 3 euros for showing up. The exchange rate

is 1.5 e for 100 tokens. This means that your total earnings will be the total amount of tokens

you have multiplied by 0.015. You will be told how much you have earned at the end of the

experiment.

NOTE

You will be told the amount to be deducted from your earnings as a result of the pairs

choosing Alternative B at the end of the experiment, when the 20 rounds are over.

40

6 Appendix 2Session Group NoBribe+A NoBribe+B PO Rejects PO Accepts+A PO Accepts+B TotalFemale F ­ Male PO 1 18 1 1 0 0 20(fm) 2 10 2 1 3 4 20

3 19 1 0 0 0 204 14 0 5 1 0 205 18 0 0 2 0 206 19 0 0 1 0 207 20 0 0 0 0 208 10 8 1 0 1 209 20 0 0 0 0 2010 7 2 1 5 5 2011 20 0 0 0 0 2012 7 0 0 6 7 2013 19 1 0 0 0 20

Male F ­ Female PO 1 15 2 0 2 1 20(mf) 2 19 0 0 1 0 20

3 20 0 0 0 0 204 3 5 2 8 2 205 1 0 3 5 11 206 8 3 1 4 4 207 3 2 0 4 11 208 3 0 0 17 0 209 17 1 0 2 0 2010 12 2 0 3 3 2011 18 2 0 0 0 2012 20 0 0 0 0 20

Male F ­ Male PO 1 2 7 0 1 10 20(mm) 2 18 1 1 0 0 20

3 18 0 2 0 0 204 20 0 0 0 0 205 18 2 0 0 0 206 8 4 2 1 5 207 15 0 0 1 4 208 17 2 1 0 0 209 16 3 1 0 0 2010 3 2 1 6 8 2011 3 0 0 1 16 2012 19 0 0 1 0 2013 15 0 0 2 3 20

Female F ­ Female PO 1 3 2 11 0 4 20(ff) 2 10 0 4 5 1 20

3 20 0 0 0 0 204 20 0 0 0 0 205 20 0 0 0 0 206 19 0 1 0 0 207 19 0 1 0 0 208 19 0 0 1 0 209 20 0 0 0 0 2010 17 0 0 3 0 2011 19 0 1 0 0 2012 15 2 0 3 0 2013 18 0 1 1 0 20

Table A5: Data by group and session

41

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Period

Freq

uenc

y of

 B c

hoic

es

Female F ­ Male PO Male F ­ Female PO Male F ­ Male PO Female F ­ Female PO

Figure A1: Frequency of B choices by period and session

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1 2 3 4 5 6 7 8 9 10

Bribe amount

Freq

uenc

y of

 B c

hoic

es

Female F ­ Male PO Male F ­ Female PO Male F ­ Male PO Female F ­ Female PO

Figure A2: Frequency of B choices by amount of the bribe and session

42